Practical guide
AI fluency for strategy analysts: test a weighted option model
Build a fictional weighted option model and sensitivity check without presenting illustrative weights as objective truth.
Strategy analysts can use AI to calculate and challenge an option model while preserving the origin of every score and weight. This fictional exercise produces a transparent sensitivity table, not a forecast, investment advice or proof that the highest score is best.
Define options and criteria
A model should not hide a vague choice behind arithmetic.
The fictional choice is between a focused pilot, a broad pilot and no pilot. Criteria are learning value, operational load and reversibility. Define each criterion and direction before scoring.
Ask AI to detect criteria that may overlap. Learning value is not predicted revenue, and operational load is not total cost. Keep criteria narrow enough to score from the supplied exercise.
Mark weights as judgments
Weights express a decision preference, not an observed fact.
Use illustrative weights of fifty percent learning, thirty percent load and twenty percent reversibility. Record who would need to approve real weights and why these numbers are used only to demonstrate the method.
Do not cite the weights as a benchmark or optimize them to produce a preferred winner. Confirm that they sum to one hundred percent and document the scoring scale.
Calculate with a visible trail
Every total should be reproducible from the component scores.
On a five-point scale, the focused pilot scores 4, 4 and 5; the broad pilot scores 5, 2 and 2. After reversing the load scale so higher is better, calculate each weighted total and retain the formula beside the values.
Use AI or a spreadsheet to calculate, then recompute manually. Do not round intermediate values in a way that changes ordering, and flag missing scores rather than substituting zero.
Test the reversal threshold
Sensitivity matters when a reasonable preference change alters the ranking.
Reduce learning weight and increase reversibility while keeping the total at one hundred percent. Record whether the ranking changes and identify the approximate threshold at which it does. The threshold describes this fictional model only.
If small changes reverse the result, report the choice as sensitive instead of asserting a winner. Also test one plausible score change because score judgments can matter as much as weights.
Deliver a decision aid with caveats
The model should expose disagreement rather than launder it.
Include option definitions, criteria, weight owners, scores, formulas, sensitivity cases and unresolved evidence. Present the no-pilot option even if its learning score is low, because it preserves a real choice.
Review arithmetic, provenance and sensitivity. Decision science may evaluate broader uncertainty; this page offers a compact strategy model whose illustrative numbers remain visibly contestable.
Trace a disagreement about one score
A score should retain the evidence and judgment that produced it.
Suppose one fictional owner scores broad-pilot reversibility as two while another argues for three because the scope can be reduced. Record both rationales and calculate the model under each score. Do not average the judgments unless the decision owner approves that rule, and do not select the score that preserves a preferred ranking.
Ask what observable evidence would resolve the disagreement, such as a documented exit condition. If the choice is unchanged under both scores, report that local robustness while retaining broader caveats. If it changes, show the decision owner exactly which judgment controls the reversal.
Sources and scope
NIST: AI risk management. Skills England: workplace AI foundations.
These references provide background, not validation or endorsement of this exercise. The case details, calculations and suggested review questions are original instructional material. Use them to discuss observable work, not to infer customer outcomes, professional credentials or performance in every setting. Before adapting the exercise, confirm the relevant facts, approved tools, data permissions and decision owners. If you change the case, revisit the expected answers and checks as well. These examples describe practice tasks, not a promise that a particular product includes the fictional features.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England